THE AMPSA DESIGN PHILOSOPHY

Transcription

1 THE AMPSA DESIGN PHILOSOPHY Pieter L.D. Abrie, Ampsa (PTY) Ltd. 1. INTRODUCTION The Ampsa Amplifier Design Wizard (ADW) is the result of thirty years of intensive development. It is mainly used to design linear small-signal RF and microave amplifiers and linear RF and microave poer amplifiers (The transistors used in the ADW are generally biased for class A, class AB or class B operation), but it also has useful oscillator analysis features. The impedance-matching capabilities provided [1] can be used for any application if the matching problem to be solved is knon (load-pull data, etc.). Harmonic control features for high-efficiency amplifiers (class J [6], F, etc.) are also provided in the latest version. Amplifier design ith the ADW is a structured process. The stages in the amplifier are designed sequentially (stage after stage). Impedance-matching and/or modification netorks (feedback and/or loading netorks added to a transistor) may be designed for each stage. Extraction of the real-frequency impedances (and the gain) required for the synthesis of the matching or modification netorks is based on constant poer contours, constant gain circles and constant noise figure circles. The extraction is done in izards and is automatic, but allos for user control. Optimization or optimization by re-synthesis is alloed at any stage during synthesis process and also hen the basic design has been completed. This structured approach drastically improves the overall productivity of the design process. In addition to speeding up the process, it also leads to greater creativity and provides much more insight in the design than the standard optimization approach. Poer control in the Amplifier Design Wizard is based on an extension of the load-line approach commonly used at RF frequencies. By shoing that it is actually the intrinsic load impedance that should be controlled and that the output poer is inherently limited by clipping of either the intrinsic output voltage or the intrinsic output current in a transistor, Steve Cripps extended the usefulness of the RF load-line approach to microave frequencies [6 7]. This approach as generalized in [2 3] by the introduction of the poer parameters. These parameters map the intrinsic current and voltage of a transistor to the external voltages of the embedding circuit. This allos the intrinsic load line to be controlled (at the fundamental, as ell as at harmonic frequencies) even hen loading netorks, impedance-matching netorks or feedback is added to a transistor. The alloable load-line area on the intrinsic I/V-plane is also defined by four boundary lines, instead of only the maximum current and voltage. This provides a better approximation to reality and also introduces the option to control on hich boundary the current of voltage clipping ill start. The netorks designed ith the Amplifier Design Wizard are practical and are adequate for surface-mount and chip-and-ire applications, as ell as for MMICs. The netorks could consist of transmission lines, single-layer parallel-plate capacitors, square-spiral inductors, bond ires and/or lumped components. Parasitic components can be specified for the lumped components and alloance is made for mounting pads too. The discontinuity effects associated ith any transmission lines used are automatically compensated. The discontinuity models used

2 can also be customized to improve the accuracy in lo-impedance circuits (high poer) or at millimeter-ave frequencies. The artork for an ADW circuit is created automatically from the schematic. Usercontrollable constraints are built into the synthesis process to ensure that the artork can be created for the netorks synthesized, and that the microstrip performance ill be close to the target electrical performance. The artork of an ADW circuit can be refined by curving, bending or meandering selected lines. When this is done, the line lengths are adjusted automatically to keep the electrical design the same. Commands are also provided to modify stubs or to replace open-ended stubs ith equivalent main-line sections. The schematic can also be refined on completion of a synthesis step. Inductors can be replaced ith hair-pin inductors, bond ires (single or double bond ires are alloed), square spiral inductors or solenoidal coils, hile capacitors can be replaced ith single-layer parallel-plate capacitors or overlay capacitors. If necessary, the circuit as designed up to that point can be optimized to restore the target performance before proceeding ith the synthesis process. Impedance-matching netork synthesis is based on the transformation-q approach outlined in [1 3]. Real-frequency specifications [8] are used to define the matching problem to be solved. The transducer poer gain of the matching netork in the passband, and the source or load reactance at harmonic frequencies (narroband netorks), can be controlled. Lumped, distributed and mixed lumped/distributed matching netorks can be synthesized. Pads and parasitic components can also be specified for the lumped elements and are taken into account in the synthesis process. Multiple solutions are provided to each matching problem solved. The sensitivity of each netork presented is also evaluated and serves as a guide to the designer in selecting the best solution to the matching problem. The synthesis of the modification netorks (feedback and/or frequency-selective resistive loading netorks) is based on extensions of [9]. These netorks are essential in most amplifiers and serve to level the gain, provide the degree of stability required and reduce the gain-bandidth constraints of the matching problems to be solved. It is useful to vie adding modification sections to a transistor as a pre-conditioning step before the associated matching problems are solved. The Amplifier Design Wizard is typically used as a front-end to one of the generalpurpose microave circuit simulators available on the market. The ADW artork can currently be exported as Microave Office scripts, as native Sonnet Softare files or in DXF format. A CST Microave Studio technology file is also created hen the artork is exported in DXF format. This allos for extruding the DXF layers at the required elevation ith the required thickness in Microave Studio. Footprints are also created for any bond ires used (the actual bond ire is not created yet). The amplifiers designed ith the Amplifier Design Wizard are usually good enough to only require fine tuning in a general-purpose simulator. This applies to class A, as ell as class AB and class B amplifiers hen the 2 nd harmonic intrinsic output voltage can be neglected. (It should be noted that the optimum fundamental frequency poer load line for a class B stage is generally very similar to that of a class A stage at the same DC operating point.) The matching netorks for high-efficiency amplifiers (like Class F and continuous class F amplifiers [10], as ell as Doherty amplifiers [11]) can also be designed in the ADW, but the simulation of the amplifier must be done externally. Depending on the design, measured or simulated load-pull data may be required to define the associated matching problems.

3 2. POWER CAPABILITY An important poer feature in the Amplifier Design Wizard is that non-linear transistor models are not used in it. (A non-linear model may be used in another tool to generate the S-parameter, noise parameter and/or load-pull data required in the ADW.) In order to control the poer performance, linear models must be fitted to the S-parameter and the noise-parameter data associated ith the desired DC operating points (not necessarily the bias point) of the different transistors used. The required models can be fitted in the Device-Modification section of the Amplifier Design Wizard. Initialization and optimization features are provided for this purpose. The S-parameters associated ith the model should fit the measured data tightly. I/V-curve boundary lines must also be specified for each transistor used. These define boundaries on the fundamental tone intrinsic output current and voltage of each transistor. The parameters of the I/V-curve boundary lines can be established by using measured dynamic (preferable) or static I/V-curve measurements. When I/V-curve data is not available, the relevant parameters can be estimated and adjusted based on any poer data information (like P 1dB ) available. Poer parameters [2 3] are used to estimate the maximum linear output poer (output poer just before the onset of clipping of the intrinsic output current and/or voltage) obtainable from each transistor. The optimum poer terminations for the (modified) transistor can be calculated and displayed. Contours of constant maximum linear output poer can also be displayed ith the optimum terminations. Ideally, the optimum poer terminations should be compared ith actual load-pull measurements to ensure that the model extracted is adequate. Note that in GaAs FETs and bipolar transistors, the maximum linear output poer usually corresponds closely to the 1dB compression point (P 1dB ). When the poer performance of a multi-stage amplifier is calculated, the influence of each transistor on the poer performance is established. This is done by assuming that clipping occurs in only one transistor at a time, ith the other transistors ideal. The output poer of each stage is referenced to the load. (That is, the output poer is increased ith the operating poer gain of the stages beteen the transistor of interest and the actual load termination.) Note that poer margins can be specified for the driver stages during optimization. This ill ensure that the output poer ill be limited mainly by the load stage. The poer of each stage can also be controlled to be higher than a target level, or to be ithin a target indo. The latter option is important hen the output poer is required to be constant over frequency (limiting amplifiers). An estimate of the difference beteen the maximum linear output poer and the saturated output poer can also be specified for each transistor hen the model is created. This specification is used ith the small-signal gain to define a saturation curve [4]. This saturation curve is used to estimate the actual output poer and the compression depth of each transistor for a given level of the input signal. This feature allos for establishing proper poer levels at the different points in the amplifier chain, and is also useful to ensure that a transistor in the amplifier chain is not overdriven. It also allos for indirect control over the harmonic and intermodulation distortion generated. All of these poer features allo for fast simulation and optimization of cascade amplifiers ith several stages. It also allos poer amplifiers to be designed hen non-linear

4 transistor models are not available yet. It should also be noted that non-linear models are not alays very accurate and better results may be obtained ith the ADW approach. 3. THE CONVENTIONAL APPROACH TO DESIGNING RF AND MICROWAVE CIRCUITS The prevalent approach to the design of RF and microave circuits today is still the optimization approach. An initial solution is obtained by some means or other, and this solution is optimized (improved by using optimization techniques) ith a general-purpose RF and microave circuit simulator. This approach has the folloing disadvantages: The optimization as provided in most circuit simulators is limited to a specific predefined topology. A circuit (amplifier) is usually optimized as a hole. It is not systematically put together (synthesized) in a step by step process. In some cases, a single-frequency circuit is synthesized systematically for use as the initial solution in a broadband design, but there is no guarantee that the specific circuit chosen is the optimum, or even a good choice, for the ide-band problem. Optimization problems are subject to the phenomenon of local minima and, therefore, the initial values assigned to the components. Even ith a fixed topology, the chances of finding the global optimum to a practical, non-trivial microave optimization problem are often poor. The optimum targets for the problem to be optimized are usually not knon, at least not initially. Because only a single solution is optimized, no perspective is obtained on the problem solved. While this approach is certainly orkable, it is slo, depends heavily on the experience of the designer (or some other source), and usually yields inferior results. The requirement of initial solutions is also a major draback. 4. DESIGNING CIRCUITS BY DOING SYNTHESIS-BASED SYSTEMATIC SEARCHES In the ADW the basic point of departure is that one cannot rely only on optimization techniques to find close-to-optimum solutions. A solution to this problem is to find solutions by doing

5 systematic searches. A simple grid search is, hoever, not the anser. In order to do the required search efficiently, the search must, in the first place, be done on the correct parameters and, secondly, it should be synthesis based. A number of the best solutions found in the systematic search are then optimized, after hich the user can choose the best solution from the potential solutions presented. Note that the systematic searches also eliminate the initialization problem. 4.1 IMPEDANCE MATCHING WITH THE ADW When impedance-matching netorks are synthesized, the transformation Qs, as defined in [1 3], ere found to be excellent parameters for a systematic search. Similar to the Q of a circuit, these Qs are inherently constrained in range, and the step size can be chosen realistically ithout significantly reducing the probability of finding close-to-optimum solutions. The last to transformation Qs in a netork are constrained further by synthesizing the netork to approximate the specified gain at, at least, one of the frequencies in the passband (a gain indo is set for this purpose). The efficiency of the search can also be increased by doing a coarser main search combined ith finer searches around a number of the best solutions (usually 10 to 25) obtained in the main search. The best solutions obtained are then optimized ith a steepest-gradient optimization routine (random optimization is, in principle, not used). It is important to realize that the searches done are topology independent. It is, hoever, a simple matter to constrain the search on the transformation Qs to result in a ide variety of practical topology constraints. Some of the constraints hich are alloed in the ADW are to limit solutions to: Lo-pass form. High-pass form. Netorks ithout any series capacitors. Netorks ithout any shunt inductors. Netorks suitable for interstage biasing purposes (4 or more elements are required). Constraints on the characteristic impedances, the line lengths and the values of any lumped components ere also implemented. The option to replace shunt capacitors ith shunt overlay capacitors is also provided. The impedance-matching modules can synthesize Lumped-element matching netorks. Commensurate distributed matching netorks (The line lengths are fixed and the line idths are used as variables). Non-commensurate distributed matching netorks (The line idths are fixed and the line lengths are used as variables). Mixed lumped/distributed matching netorks (Lumped elements are used to minimize the line lengths).

6 Note that the commensurate matching algorithm implemented as generalized so that different lengths can be used for the main-line sections, the open-ended stubs and the shortcircuited stubs. By using short stubs, the option to replace any stub ith an equivalent stub, or a lumped component ith pads, is introduced (if the line length is short, the tangent function is essentially linear). Matching netorks for poer transistors are often designed ithout using any stubs. When non-commensurate impedance-matching netorks are synthesized this can be done by using the folloing options: Choose the topology option not to have any shunt inductors (short-circuited stubs). Choose the option to replace any open-ended stubs ith double stubs. Choose the option to replace double stubs ith stepped main-line sections. When commensurate impedance-matching netork are synthesized, the lo-pass option can be used, and the line length to be used for open-ended stubs can be set to be very short. The constraints on the alloable idths of these stubs should be set to force the stubs to be narro. Note that a schematic command to transform an open-ended stub to a stepped main-line section is provided in the Analysis Module, ith various other transformation options. When mixed lumped/distributed matching netorks are synthesized, the line lengths are reduced by using lumped elements. Constraints can also be imposed on the values of the lumped elements. (From the viepoint of accuracy, a short line or stub is preferable to a lumped component.) The pad size to be used ith each type of lumped component (series or shunt capacitors, or series or shunt inductors) must be specified by the user. In order to minimize the parasitic effects introduced, any pads specified should be as small as possible. All of the matching netorks are synthesized to approximate the specified transducer poer gain versus frequency response beteen to specified complex terminations. The specifications must be made in real-frequency format (terminations and gain specified at a number of frequencies in the passband of interest). No equivalent circuits or an analytical expression for the gain response is required, or used. The gain at each frequency can be specified to be less than, greater than or equal to the gain value specified. The less than option is useful hen rejection is required. In order to evaluate the relative sensitivity of the different solutions to component changes, a orst-case tolerance analysis is done on each netork in the results section of the relevant impedance matching module. The default tolerance value is 1%. The choice of the specific solution to be used is left to the user. Having several solutions to choose from (different topologies and different element values) is an advantage from a manufacturing viepoint and also provides perspective on the problem being solved. The option to create an ADW circuit file for the solution chosen, or to insert the solution chosen at a marked position in a previously created circuit file, is provided in the results section of the impedance-matching module. Microstrip capability The ADW synthesizes netorks electrically, that is, in terms of inductance, capacitance, resistance, characteristic impedance and electrical line length. If microstrip or stripline solutions

7 are required, these electrical solutions must be transformed and optimized before they are of practical use. A distributed ADW circuit can be transformed automatically into microstrip or stripline form, ith the option to adjust the line lengths to compensate for the shifts in the reference planes caused by step, tee and cross junction discontinuities. Step junctions can also be tapered to reduce the associated discontinuity effects. If suitable values ere chosen for the characteristic impedances and line lengths, the performance of the transformed circuit usually corresponds closely to that of the original circuit, even hen complete amplifier circuits are transformed. Note that the characteristic impedances should be chosen to minimize any transformer effects or reactive loading effects at the different junctions. The information required to make an appropriate choice is provided hen required. It should be realized that the discontinuity effects decrease as the substrate height is decreased. Hoever, hen lines are too narro, the conductor losses ill be unacceptable. An ADW circuit can be exported in Super Compact TM or Touchstone TM netlist format, or as a Microave Office TM script (This script can be run in Microave Office TM to create the schematic and the associated artork). Native Sonnet Softare files (planar EM simulation) can also be created for the ADW artork. DXF files can also be exported by the ADW and these can be imported into an EM simulator. Many features ere implemented to simplify the process of exporting the ADW artork to CST s Microave Studio (3D EM simulator). Note that a CST technology file is also created hen the ADW artork is exported in DXF format. This technology file is used in Microave Studio to extrude each DXF layer at the correct elevation ith the correct height. Layer mapping is alloed in the ADW for single-layer capacitors, bond ires and vias, for different substrates and also for mounting transistor dies on ridges (chip-and-ire applications). Overlay capacitors ith a centered via, or ith a single or to offset vias, are also alloed. The ADW has been set up to handle chip-and-ire circuits, as ell as basic MMIC circuits ell. Single or double metallization, air-bridges, a capacitor dielectric layer and vias can be used hen MMIC circuits are designed in the ADW. Models for single-layer capacitors, square spiral inductors, single and double bond ires and solenoidal coils are also provided. The parameters in these models can be optimized by using a 2D or 3D EM simulator. If the circuit or sub-circuit exported is optimized externally, the to-port parameters of the optimized netork can be imported into the ADW (Touchstone *.s2p format). This approach is also folloed hen circuits are realized on media other than microstrip or stripline (co-planar, suspended substrate or aveguide circuits are not yet supported in the ADW). Similarly, any components not presently supported in the ADW (like radial stubs, interdigital capacitors, circular spiral inductors, etc.) can be designed ith another softare tool, after hich the associated S-parameters (and noise parameters) can be imported into the ADW. Note that an ADW circuit can be exported in electrical or geometrical format. The electrical circuit description can be used as target for any external microstrip/stripline optimization required. Depending on the substrate height and the impedance levels, ADW microstrip or stripline circuits are usually adequate up to at least 12GHz. Hoever, customization features ere implemented to allo accurate results even at millimeter-ave frequencies, and also for lo-impedance poer amplifier circuits. These features are available in the professional version of the ADW.

8 The Equivalent Passive Problem The best ay to synthesize an amplifier is to control the available poer gain and the noise figure, or the operating poer gain and the output poer, of each stage in the amplifier. The design cycle should be started at the source side hen a lo-noise amplifier is designed and at the load side hen a poer amplifier is designed. When a high-dynamic range amplifier is designed, the lo noise section and the poer section can be designed separately and can then be joined by using an interstage matching netork. If this approach is folloed, the input port or the output port of each (modified) transistor used must be terminated in an impedance hich lies on or near to the circumference of the target constant gain circle or constant noise figure circle or constant poer contour, at each frequency of interest. An impedance-matching netork must, therefore, be synthesized to transform the existing impedances at the port considered to those required at the different frequencies of interest. If the transistor is inherently stable or as compensated to be so [1 3], it is possible to transform each of the circle problems (constant gain or constant noise figure circles) to a passive impedance-matching problem hich is exactly equivalent (Inherent stability is not required hen the noise figure is controlled). To understand ho this is possible, it is only necessary to realize that the problem of matching a passive source to a passive load is also a circle problem hen the required transducer poer gain is less than unity [1 3]. The active problem, therefore, reduces to finding the passive problem for hich the circle required is exactly the same as the target constant noise figure or constant gain circle, at each frequency of interest. In those cases here inherent stability as not established or hen the output poer is controlled (elliptic contours), the optimum point on the relevant circle or poer contour can be selected to complete the specifications for the equivalent passive problem. The option to enforce such a point-match is also provided in the inherently stable case. Note that, hen a circlematch is possible, a point-match should only be enforced if the performance is only acceptable over a small segment of the relevant circle. The option to optimize the matching netorks synthesized to provide a better fit to the targeted active performance is provided in the results section of the Impedance-Matching Module. 4.2 DEVICE MODIFICATION The capability to solve impedance-matching problems and the concept of the equivalent passive problem are critical factors in establishing a general amplifier synthesis capability. Hoever, it ill be found that these alone do not bring one very far. The folloing problems, all of hich are associated ith the transistor(s) to be used, still need to be solved: Stability outside the passband can be a major problem. Synthesizing a netork that ill provide the required performance inside the passband and that ill ensure stability outside the passband can be difficult. When it can be done, unnecessary constraints are imposed on the performance inside the passband.

9 All non-trivial impedance-matching problems have inherent gain-bandidth constraints. If these constraints are too severe and cannot be reduced, the bandidth can only be increased by degrading the associated VSWR (that is, hen a lossless matching netork is used). Transistors have inherent gain slopes at higher frequencies. While these gain slopes can be removed ith lossless netorks ith suitable positive slopes, the sensitivity to any variation in the component values is frequently a problem. The optimum noise figure or optimum poer termination for a transistor is usually very different from the optimum gain termination. The anser to these problems is to add frequency-selective, resistive, feedback and/or loading netorks to the transistor. In doing so, the transistor is modified to be more suitable to the application at hand. These netorks ill be referred to as device modification netorks, and each resistive section ill be referred to as a modification section. To series of modification techniques ere originally developed. Only the double-section modification technique is implemented in recent versions of the ADW. The main reason for this is that a single modification section rarely provides the performance required, and hen it is the best option, the best double-section netork ill effectively reduce to a single section GAIN SLOPE CONTROL WITH LOSSY SECTIONS In the first series of modification techniques developed, a table of the components required to provide a specified slope in the maximum available poer gain (MAG) or the maximum stable poer gain (MSG) is created. (To understand hy the MAG is targeted, is it necessary to understand that a flat gain response and lo input and output VSWRs can only be obtained ith lossless matching netorks if the MAG of the transistor has been leveled first). Tables can be created for Voltage-shunt feedback sections Current-series feedback sections Shunt loading sections Series loading sections In each case, the gain slope is controlled ith a resistor used in combination ith some lossless component(s). (In the simplest case, a resistor is used in parallel ith a capacitor or in series ith an inductor). By using a resistive netork to reduce the gain slope, the stability and the VSWRs may be improved simultaneously. In general, sensitivity is also not a problem. This follos because these netorks usually have Qs hich are lo in comparison ith those required from an equivalent lossless netork.

10 Note that the MSG can only be controlled ith a feedback section. If the slope in the MAG is set to be equal to that in the MSG, tables of the components required to just stabilize the transistor at different gain levels can be created. Different gain-slope control sections can be combined sequentially. Some fraction of the gain slope can first be removed ith one section, after hich a second section can be used to level the gain. The first section can be used to control the MSG or the MAG, hile the second is usually used to level the MAG. Note that if feedback sections, as ell as loading sections are used, the feedback must be applied before the loading. Better VSWRs can often be obtained hen to sections are combined to reduce, or to level, the slope in the MAG. At least to modification sections are usually required for good results THE DOUBLE-SECTION MODIFICATION TECHNIQUES In the second series of modification techniques any to of the different modification sections (voltage-shunt feedback, current-series feedback, shunt loading and series loading) are combined automatically to level the gain of the stage synthesized. While only the MAG or the MSG could be controlled ith one of the single-section modification netorks, any of the folloing gain functions can be controlled ith a double-section modification netork: The MAG. The MSG. The transducer poer gain (G T ). The available poer gain (G a ). The operating poer gain (G ). The available poer gain associated ith the optimum noise figure of the modified transistor (G anopt ). Note that the performance of any stages synthesized previously (multistage amplifier) are taken into account hen the gain is controlled in the ADW. The netorks synthesized are optimized first before the solutions are presented to the user. Optimization is in terms of an error function, through hich control is provided over the folloing parameters: The average poer gain over the passband. The gain ripple over the passband. The highest input VSWR and output VSWR in the passband. The highest noise figure in the passband. The degree of difficulty of the noise match, expressed as a VSWR. The loest Rollette stability factor in the passband or over the complete frequency range over hich S-parameter data ere provided for the transistor.

11 The load and source stability factors (LSF and SSF; these factors are generalizations of µ and µ-prime [5]). The maximum linear output poer. The output poer can be optimized to be higher than a critical level or to be ithin set boundaries. Note that the VSWRs targeted could be actual VSWRs (like the final input VSWRs) or could be VSWRs used to indicate the degree of difficulty of the impedance-matching problem to be solved. To appreciate this, consider the case here the MAG is controlled. If perfect matching netorks could be designed, the input VSWRs, as ell as the output VSWRs, ould be unity hen the matching netorks are in place. In the ADW, the poorest VSWR in the passband before a matching netork is added to a circuit is used as an indication of the degree of difficulty of the associated matching problem. The VSWR is calculated in terms of the actual complex source or load impedance in place (Z 0 ) before the matching netork is added. It is important to realize that hen the MAG, G a, G or G anopt is controlled, a conjugate match at the other port of the modified transistor is implied. This implies that a matching netork is required at that port before the performance targeted can be realized. No matching netorks are required hen the transducer poer gain (G T ) is controlled. When G anopt is controlled, an optimum noise match is required on the input side and a conjugate match on the output side. Contrary to popular believe, excellent lo-noise stages can usually be synthesized by using resistive sections. Such sections should, hoever, be used on the output side of the transistor. The best lo-noise results are usually obtained by synthesizing the modification section to control G anopt or the MAG. Similarly, modification netorks are usually added on the input side of the transistor hen the output poer should be maximized. Similar to impedance-matching netorks, the double-section modification netorks are synthesized by doing a synthesized-based systematic search. The search is done over all netorks hich ill provide the specified gain and input or output VSWR, at at least one of the passband frequencies. A number of the best solutions (up to 45) are stored and optimized ith a steepest-gradient optimization technique. The search can be global, can be constrained to not include some modification sections (like voltage-shunt or current-series feedback, or resistive loading on the input or output side), or it can be done for a specific modification section. As in the case of impedance matching netorks, the choice of the specific modification netork to be used is left to the user. In earlier versions of the ADW, no provision as made for connecting lines or for the pads required for the lumped components hen modification netorks ere synthesized. This as found to be a major draback and as corrected by adding an optimization phase to the synthesis cycle. Because of the optimization step, one-port parasitic components are also alloed for the lumped components. Note that unaccounted parasitic effects and phase shifts can have a disastrous effect on the expected performance of a voltage-shunt feedback loop.

12 5. STABILITY CONSIDERATIONS A complete picture of the relative stability of a (modified) transistor is not provided by the k factor (Rollette stability factor) typically used. One of the reasons for this is that small changes inside the circuit can cause large changes in k. It has been found that the loop gain associated ith each feedback loop should be evaluated too. Although the loop gain is dependent on the terminations used, it provides the information and the control required to handle most stability problems. The loop gain of a circuit can be evaluated in the Analysis Module of the ADW. Reflection analysis features are also provided in the Analysis Module. Note that ignoring the information provided by the k factor is not advocated. Any potentially unstable to-port (k < 1) can be stabilized by loading its input and/or output terminals ith series or shunt resistance (Loading on both sides is sometimes required). Stabilization is sometimes also possible ith a voltage-shunt feedback resistor. Valuable insight into the degree of instability is often gained from the values of the stabilizing resistance required. Tables of the resistance required are, therefore, provided in the Device-Modification Module and in the Analysis Module. The performance loss associated ith inherent stability can sometimes not be tolerated. When conditional stability is sufficient, the µ and µ-prime stability factors [5] are useful. These factors can be used to establish the change in the load termination (µ) or the source termination µ-prime (expressed as VSWRs), that can be tolerated before oscillations become possible (That is, oscillations ill not be possible as long as the VSWRs are smaller than some value). They ork ell hen the circuit terminations are purely resistive and only lossless elements are used to connect the terminations to the transistor. The load stability factor (LSF) and the source stability factor (SSF) used in the ADW are generalizations of µ and µ-prime. They allo calculation of the VSWRs that can be tolerated before oscillations become possible in terms of any arbitrary termination (and, therefore, also the actual termination) at the transistor port of interest. When they are calculated for a complete circuit ith purely resistive terminations, the LSF and SSF values calculated are equal to µ and µ-prime, respectively. 6. THE CIRCLE AND POWER MODULES The impedance-matching module of the ADW as developed to synthesize solutions to passive impedance-matching problems. Active matching problems are converted to equivalent passive impedance-matching problems hen the ADW CIR or CIL Wizards are used. Any one of the folloing parameters can be set as target in these izards: The operating poer gain (G ). The available poer gain (G a ). The transducer poer gain (G T ). The noise figure (F). The maximum linear output poer (P out ). The effective maximum linear output poer (P out -P in ).

13 The specifications set for the stage designed correspond to one of the folloing: A set of constant gain circles. A set of constant noise figure circles. A set of constant maximum linear output poer (P out ) or effective maximum linear output poer (P out -P in ) contours. A Summary Table is displayed directly after the specifications for the performance required are made. It summarizes the potential performance on the circles or contours targeted. The variation of the best and the orst performance on each circle or contour, over the passband of interest, is also displayed. Note that the circles or contours of interest can be displayed graphically too. The option to allo a circle match or to force a point match, at each of the passband frequencies of interest, should be made after selecting the circles or contours to be used. If a point match is required, the optimum point on the relevant circle or contour must be selected. The default optimum point is selected by using an error function. The parameters of this error function can be modified by the user hen the Summary Table is displayed. To allo evaluation of the performance around each circle or contour, a table listing the performance at different angles around the circle or contour is displayed. Zoom capabilities are provided in these tables in order to allo displaying of the performance at any point on the circle or contour of interest. The optimum point on the circle or contour is high-lighted in the table. The folloing parameters are listed in these tables: The poer gain (G a, G or G T ) on a constant noise figure circle or a constant output poer contour, or the noise figure (F) around a constant gain circle or a constant output poer contour. When the operating poer gain (G ) or the output poer (P out ) is controlled, the noise figure associated ith a conjugate match on the input side, or ith the termination currently in place on the input side, is calculated. The degree of difficulty of the associated impedance-matching problem expressed as a VSWR. The required impedance is compared ith the impedance actually in place hen this is done. The Sterne stability factor ith the stabilizing influence of the terminations taken into account. The tunability factor is a measure of the influence that any impedance changes (tuning) on the output or input side of the transistor ill have on the impedance at the other side. When the input and the output sides of a transistor are perfectly isolated (unilateral case), the tunability factor is zero. A tunability factor of less than 0.3 is usually acceptable. Note that poor tunability can be an advantage hen the input VSWR of a lo-noise stage or the output VSWR of a poer stage is controlled by the termination on the other side of the (modified) transistor (indirect VSWR control via s 12 ).

14 Worst-case sensitivity factors associated ith variations in the controlling admittance. A default tolerance of 1% in the controlling admittance is assumed hen these sensitivity factors are calculated. Sensitivity factors are calculated for the noise figure (δ n ), as ell as the gain (δ g ). The available poer gain (G a ) or the operating poer gain (G ) specified can only be extracted as actual gain if the output impedance (G a ) or input impedance (G ) of the transistor is conjugately matched to the load (G a ) or source (G ). Assuming that this ill (and can) be done, an output VSWR (G a ) or input VSWR (G ) of unity is, therefore, expected after matching. Any tolerance in the controlling admittance presented to the input terminals (G a ) or output terminals (G ) of the (modified) transistor ill degrade this match. The orst-case deviation in the expected VSWR (δ vs ) is calculated and listed ith the other sensitivity factors. The tolerance circle used for the gain or noise figure sensitivity factor is also used for the VSWR sensitivity factor. A data file for the equivalent passive impedance-matching problem to be solved is created automatically after the option to modify the default name supplied for it has been used. The Impedance-Matching Module can then be launched to synthesize solutions to the defined matching problem. Note that the problem is solved in terms of the equivalent passive problem. The option to optimize the netorks synthesized in terms of the active performance is provided in the results section of the Impedance-Matching Module. 7. ARTWORK GENERATION As described above, the basic artork for an ADW circuit is generated automatically from the electrical circuit description. Extensive graphical manipulation of the microstrip artork is possible. This includes the folloing: Flipping stubs around. Adjusting the gap spacing or offsets vector used for a lumped component. Adjusting the dimensions of a line. Bending a line (optimum miter). Curving a line. Meandering a line. Tapering a step junction. Closing any misaligned voltage-shunt feedback loop or series block loop (to cascade netorks connected in parallel) automatically. A suitable component must be selected before the loop is closed.

15 When a line is bent or curved or meandered, the physical length of the line is adjusted to keep the electrical line length unchanged. Note that physical line lengths may also be adjusted to compensate for step junction, T-junction or cross junction discontinuity effects. The adjustments are made to get the performance of the microstrip circuit as close as possible to that of the electrical circuit. With the exception of nodal blocks and some components, ADW circuits can be exported to Microave Office. The script created by the ADW can be executed in Microave Office to create the schematic of the circuit, as ell as the associated artork. Any residual discontinuity effects can then be removed by optimization before the circuit is manufactured. Any elements not available in the ADW can also be added at this stage. Sonnet Softare files, as ell as DXF or HPGL files can also be created for the ADW microstrip artork. The Sonnet files and the DXF artork can be used to verify or optimize the performance of the different microstrip netorks designed in the ADW, or can be used to create simulation data for fitting models to passive components like single-layer capacitors, bond ires, spiral inductors and solenoidal coils. It is a good idea to export individual netorks for simulation in Sonnet s EM. Adjustments can be made to correct for discrepancies as the design proceeds and if the discrepancy is severe, a different netork should be considered. 8. OPTIMIZATION AND OPTIMIZATION BY RE-SYNTHESIS One of the final steps in an ADW design cycle is usually to optimize the completed design in the Analysis Module. (Note that, after the optimization, it is important to use the loop gain analysis feature to check the stability of each stage in the amplifier before the design is exported for further processing ith another tool.) An Optimization Wizard is provided to set up the parameters of the error function to be used. Extensive control over the gain, the noise figure, the VSWRs, the stability factors (k, SSF and LSF), the maximum linear output poer, the maximum linear poer provided by each driver stage, and the maximum gain compression is provided. Error factors are also provided to prevent clipping of the intrinsic output voltage or current of a transistor on any of the four I/V-plane boundary lines alloed. It is also possible to control the amplification of the harmonics relative to the associated fundamental tone in ideband amplifiers. This is important in ideband amplifiers hen lo harmonic distortion is required. The topology ill usually not be changed hen a circuit is optimized directly. The Synthesis Wizards can, hoever, be used to re-synthesize any of the any of the matching or modification netorks in the amplifier. Note that the To-port Command on the Schematic Toolbar can be used to replace any of the transistors used in the circuit ith a different transistor, or the same transistor used at a different DC operating point, if necessary. The izards provided in the ADW are listed belo, ith a description of the function of each izard.

16 General Impedance-Matching Wizard The IIM, RMT, LMT and NOI commands can be activated by using this izard. The different commands are described belo: IIM This command can be used to set up an interstage impedance-matching problem. It can also be used to set up the matching problem associated ith improving the input or the output match of the circuit. The transducer poer gain for the matching problem is set up automatically to level the overall transducer poer gain of the circuit over the passband of interest. If a good match is required, the gain should be set to unity at each of the passband frequencies. This can be done in the impedance-matching section. The source terminations for the left-hand side section of the circuit (section on the input side of the insertion point) are taken to be the actual source terminations of the circuit (as defined in the terminations block of the circuit file), hile the load terminations for the right-hand side section are taken to be the actual load terminations. RMT The RMT command can also be used to set up an interstage impedance-matching problem, but instead of using the actual load terminations as load terminations for the section to the right of the insertion point (the output section), the terminations associated ith the MAG (maximum available gain) of that section is used. LMT The LMT command is similar to the RMT command, but the input side of the section to the left of the insertion point (the input section) is terminated in its MAG terminations. NOI The NOI command is used to minimize the noise figure of the circuit section to the right of the insertion point (the output section). The circuit is assumed to be terminated on both sides as specified in the circuit file. IVI Wizard This izard is used to set up an interstage impedance-matching problem to improve the input VSWR of the section to the left of the insertion point (input section) by mismatching its output impedance to the input impedance of the section to the right of the insertion point (output section). It can be used to improve the input VSWR of a lo-noise amplifier. The gain of the amplifier can usually be leveled at the same time.

17 OVI Wizard This izard is similar to the IVI Wizard and is used to set up an interstage impedance-matching problem to improve the output VSWR of the section to the right of the insertion point (the output section) by mismatching its input impedance to the output impedance of the section to the left (the input section). As in the IVI case, the gain of the amplifier can usually be leveled at the same time. CIL Wizard This izard is used to (re-)design an interstage or a load netork to control the maximum linear output poer (P out ), or the operating poer gain (G ) or the transducer poer gain (G T ) of the section to the left (on the input side) of the insertion point. Note that information on the detailed performance (efficiency, gain, poer, noise figure, stability factors, etc.) at each point on the contour is provided. If necessary, a point match can be enforced. CIR Wizard This izard is used to (re-)design an interstage or a source netork to control the noise figure (F), or the available poer gain (G a ) or the transducer poer gain (G T ) of the section to the right (on the output side) of the insertion point. Information on the detailed performance (efficiency, gain, poer, noise figure, stability factors, etc.) at each point on the contour is provided. If necessary, a point match can be enforced. MOT Wizard This izard is used to (re-)synthesize the modification netork of a transistor and/or to replace the current transistor in an amplifier chain. It sets up the information required by the Device- Modification Module to insert a ne stage (a transistor ith its modification netorks) at the insertion point.

18 9. REFERENCES 1. Pieter L.D. Abrie, The Design of Impedance-Matching Netorks for Radio-Frequency and Microave Amplifiers, Artech House, Inc., Pieter L.D. Abrie, Design of RF and Microave Amplifiers and Oscillators, Artech House Inc., Pieter L.D. Abrie, Design of RF and Microave Amplifiers and Oscillators, Second Edition, Artech House Inc., K.M. Johnson, Large Signal GaAs MESFET Oscillator Design, IEEE Trans. Microave Theory and Tech., Vol. MTT-27, No. 3, March M.L. Edards and J.H. Sinsky, A Ne Criterion for Linear 2-Port Stability Using a Single Geometrically Derived Parameter, IEEE Trans. Microave Theory Tech., Vol. MTT-40, No. 12, December S.C. Cripps, GaAs Poer Amplifier Design, Technical Notes 3.2, Paolo Alto, CA: Matcom Inc. 7. S.C. Cripps, Advanced Techniques in RF Poer Amplifier Design, Norood, MA: Artech House, J.H. Carlin, A Ne Approach to Gain-Bandidth Problems, IEEE Trans. Circuits Syst., Vol. CAS-24, April 1977, pp P.L.D. Abrie, A Series of CAD Techniques for Designing Microave Feedback Amplifiers and Simplifying the Design of Reactively Matched Single-Ended Amplifiers, IEEE MTT-S Digest, N. Tuffy, L. Guan, A. Zhu, and T. Brazil, A Simplified Broadband Methodology for Linearized High-Efficiency Continuous Class-F Poer Amplifiers, IEEE Trans. Microave Theory and Techn., Vol. 60, No. 6, June W.H. Doherty, A Ne High Efficiency Poer Amplifier for Modulated Waves, Proceedings of the IRE, Vol. 24, No. 9, September 1963, pp